IMPACT OF DE-IDENTIFICATION ON MASTER PATIENT INDEX AND DATA LINKAGES August 2020 Kathy Hines, Senior Director of Partner Operations & Data Compliance Scott Curley, Manager of Privacy & Compliance CENTER FOR HEALTH INFORMATION AND ANALYSIS
OVERVIEW
Motivation for Change Rising external cybersecurity threats to healthcare data Internal risks of accidental or intentional data exposure. Specific to the APCD – Federal Law 42 CFR Part II Impact of De-identification on MPI and Data Linkages 3 | Scott Curley, Kathy Hines| August 2020
Analytic Challenge Outright removing PII would prevent CHIA and our external community of data users from connecting health care encounters across carriers and to other datasets CHIA set an objective to dramatically decrease the risk of exposure of collected PII while retaining the ability to connect data together. Impact of De-identification on MPI and Data Linkages 4 | Scott Curley, Kathy Hines| August 2020
CHIA’s Solution 1. Software CHIA’s File Secure software is deployed to the site of data submission (insurance carriers and hospitals) that replaces key PII fields with pseudonymized equivalents 2. Internal Architecture CHIA never receives PII “in the clear” and the data is stored separately from the data warehouse and are not released to internal users or external data applicants. 3. Submission Guide Updates CHIA stopped collection of certain fields 4. Master Patient Index One ID for each person regardless of insurance carrier with the ability to link to external data Impact of De-identification on MPI and Data Linkages 5 | Scott Curley, Kathy Hines| August 2020
DE-IDENTIFICATION USING EXPERT DETERMINATION
HIPAA De-Identification Safe Harbor Expert Determination Pros Pros Methodology tailored to data Easy to implement and set in question maintain Cons Lower overall risk of re- identification 18 data elements redacted Cons or removed entirely No single method for More restrictive than implementation statistical de-identification with respect to birth dates, Routine reassessment service dates, and More restrictive than Safe geographic data Harbor with respect to some individual claim lines Impact of De-identification on MPI and Data Linkages *Slide courtesy of ONPOINT Health Data 7 | Scott Curley, Kathy Hines| August 2020
OnPoint Worked with CHIA to Define Approach Established the variables to be considered for a formal re- identification risk analysis • Catalogued all direct identifiers and quasi-identifiers Determined acceptable risk levels • Minimum cell size, maximum risk, average risk • Assumed an adversarial environment where the recipients of the data have knowledge of quasi-identifying values for the individual Established annual re-assessments * Slide courtesy of ONPOINT Health Data Impact of De-identification on MPI and Data Linkages 8 | Scott Curley, Kathy Hines| August 2020
Applied the Data Strategy The risk mitigation model was applied to multiple years of data (MA APCD Graphic should fit approximately data set years 2012 – in this space 2017) to assess the risk stability over time and project a solution for the following year. * Slide courtesy of ONPOINT Health Data Impact of De-identification on MPI and Data Linkages 9 | Scott Curley, Kathy Hines| August 2020
FILE SECURE
CHIA’s File Secure Data Submitters prepare files that include PII at their location File Secure replaces key fields with pseudonymized values (128 character length) while still at their location • Name • SSN • Full DOB (MMYYYY are left in the clear for analytics) “In the clear” versions of Name, SSN, DOB never leave the data submitter’s site Impact of De-identification on MPI and Data Linkages 11 | Scott Curley, Kathy Hines| August 2020
CHIA’s File Secure Zip code processing • Flag if invalid zip code • Retain MA Zip codes only • Map MA Zip codes to mask small areas in MA APCD State code processing • Flag if invalid state • Retain only New England and New York state codes • Map MA Zip codes to mask small areas in MA APCD File Secure encrypts the file with NIST compliant encryption before data is sent to CHIA Impact of De-identification on MPI and Data Linkages 12 | Scott Curley, Kathy Hines| August 2020
SUBMISSION GUIDE
Submission Guide Changes – Data Removal Claims • First/Last names • Social Security numbers (SSNs) • Address information Eligibility • Street/City address information • Zip code limited to 5 digits • Race/Ethnicity indicators • Disability/Marital/Student/Family size indicators • Language (list abbreviated) • Date of Death Impact of De-identification on MPI and Data Linkages 14 | Scott Curley, Kathy Hines| August 2020
Insurance Carrier Submissions CHIA Encrypted APCD Submission Files (“in Eligibility File for the clear”) Final Secure CHIA Transport Software Landing to CHIA HASH Eligibility • Zone USPS • First Name Removed: Nickname • Last Name • Street/City Table • SSN • Marital Status • NYSIIS - First • DOB • Race/Ethnicity and Last • Employee Status name Clear • Student Status • HASH • State • Date of Death Function • Zip • Remove • Insurance ID known dummy • Org ID A values • Month / Year • Re-map 1% of of Birth Medical Claims population • Gender Removed: (small ZIP PII codes) Product Provider Dental Rx Claims Insurance Carrier Site Removed: Removed: PII PII Impact of De-identification on MPI and Data Linkages 15 | Scott Curley, Kathy Hines| August 2020
MASTER PATIENT INDEX (MPI)
MPI and Record Linking CHIA creates a master patient index (MPI) using a probabilistic matching algorithm with pseudonymized identifiers. The ID connects all records that are very likely the same person and assigns them a key that is not based in any way on PII or any other attributes of a person’s data. Example of what an APCD data user might have access to • MPI – CHIA’s randomly generated unique ID for a person • MM/YYYY of birth • 5 digit ZIP code for largely populated ZIP codes CHIA has deployed a service to connect external data to APCD or Case Mix using a combination of CHIA’s File Secure software and CHIA’s probabilistic matching engine Impact of De-identification on MPI and Data Linkages 17 | Scott Curley, Kathy Hines| August 2020
CHIA Master Patient Index CHIA Data CHIA Master Data Load APCD CHIA Preparation Patient Algorithm Landing Index Hub Zone Records where: Probabilistic (MEID) Filter Known Data • Insurance ID matching method Issues • Org ID using: are the same are • considered the HASH Fname • same person. HASH Lname • HASH SSN • The last 5 valid HASH DOB • values of each Zip • input field are Gender stored to capture name changes, Links records people moving within and across etc. carriers. CHIA MPI Org ID Insurance ID First Name Last Name DOB Gender SSN ZIP Code 111111 30 BBY00002211 ABCD QRSTUVWXYZ POIUYT F HFHDSFH 02116 KDFGJKDFKFK 02461 02090 112233 22 HVD00000122 QWDD DGFGDFFGFG GFGDFF M FGDDDFG 02118 112233 30 BBY000034234 QWDD DGFGDFFGFG GFGDFF M FGDDDFG 01056 13116 01025 Impact of De-identification on MPI and Data Linkages 18 | Scott Curley, Kathy Hines| August 2020
CHIA MATCHING SERVICE
CHIA Matching Service (Master Data Management) Customer File CHIA File Linking File CHIA Secure Prepped APCD • First Name Software Algorithm • Last Name HASH High Lower • SSN • First Name Score Score • Gender Matches Matches • Last Name • DOB Scores each input • SSN • record against Zip Code Used by • DOB likely candidates in Any number of • customer Insurance ID the MPI Hub. additional match • Custom Match Study ID scenarios can be In the Clear Threshold based added and • Zip on how accurate The more complete separated from the matches need • Insurance ID the file, the better the High Score to be. • the match results Gender Matches based however not all on study need • Study ID For example: fields are needed Higher = All fields for each record for a For example: present and up to confident match Lower = SSN 1 mismatch Missing and up to 1 mismatch Customer Site CHIA SFTP to CHIA CHIA use Master Enterprise ID to identify corresponding claims, this ID is then replaced with the project’s unique Study ID and claims returned to customer Impact of De-identification on MPI and Data Linkages | Scott Curley, Kathy Hines| August 2020 20
Recommend
More recommend
